- Email: [email protected]
Estimation of cyanobacteria biovolume in water reservoirs by MERIS sensor
w a t e r r e s e a r c h 6 3 ( 2 0 1 4 ) 1 0 e2 0
Available online at www.sciencedirect.com
ScienceDirect journal homepage: www.elsevier.com/locate/watres
Estimation of cyanobacteria biovolume in water reservoirs by MERIS sensor M. Medina-Cobo a, J.A. Domínguez b, A. Quesada c, C. de Hoyos a,* ficos (CEDEX), Paseo bajo de la Virgen del Puerto 3, 28005 Madrid, Spain Centro de Estudios Hidrogra Mathematical and Fluid Physics Department, Universidad Nacional de Educacion a Distancia (UNED), Senda del Rey Street 9, 28040 Madrid, Spain c Dpt. Biology, Universidad Autonoma de Madrid, Calle Darwin 2, 28049 Madrid, Spain a
b
article info
abstract
Article history:
Planktonic cyanobacteria primarily develop in lentic water bodies, such as lakes and water
Received 22 January 2014
reservoirs. In certain instances, toxin-producing cyanobacterial populations might domi-
Received in revised form
nate the phytoplankton community. Satellite remote sensing is a useful tool for large
29 May 2014
spatial scale monitoring of cyanobacteria, and the MERIS sensor from the Envisat satellite
Accepted 2 June 2014
has taken worldwide images at a high frequency for over 10 years. This short time lapse
Available online 11 June 2014
image collection has provided an extensive record of images for the analysis of variation in the cyanobacterial communities in water reservoirs for management and scientific pur-
Keywords:
poses. The objective of this work is to determine the relationship between measured
Cyanobacteria
cyanobacterial biomass as biovolume and the estimations derived from MERIS imagery.
Biovolume
This study encompasses two independent studies relying on data from 23 water reservoirs.
Phycocyanin
First, a long-term global limnological research study was conducted that provided a field
Water reservoirs
data collection that included cyanobacterial biovolume, among other variables. Second, a
MERIS sensor
survey was conducted that applied the processed images derived from the Envisat MERIS
Time series
sensor. The chlorophyll-a (Chl a) content and phycocyanin concentration (PC) were estimated from the MERIS images. The PC estimated by remote sensing and total cyanobacterial biovolume measured from the field samples were found to be significantly correlated (R2 ¼ 0.6219; p < 0.001). No relevant differences were found among the taxonomical groups, which indicated that this tool provides accurate estimations irrespective of the cyanobacterial group. For validation, the algorithm derived from the entire dataset was applied to the MERIS image dataset of the Rosarito reservoir. An estimated cyanobacterial biovolume time series was performed and compared to the biovolume data collected in an extensive sampling schedule spanning 4 years. The results indicated a strong correlation (R2 ¼ 0.72; p < 0.001) between the measured and estimated data acquired on the same day. © 2014 Elsevier Ltd. All rights reserved.
* Corresponding author. E-mail addresses: [email protected] (M. Medina-Cobo), [email protected] (J.A. Domínguez), antonio.quesada@ uam.es (A. Quesada), [email protected] (C. de Hoyos). http://dx.doi.org/10.1016/j.watres.2014.06.001 0043-1354/© 2014 Elsevier Ltd. All rights reserved.
w a t e r r e s e a r c h 6 3 ( 2 0 1 4 ) 1 0 e2 0
1.
Introduction
Cyanobacteria are a phylum of photosynthetic prokaryotic organisms distributed worldwide (Whitton and Potts, 2002). In terms of human health risk assessment, cyanobacteria can pose a hazard when they produce cyanotoxins, a family of secondary metabolites that can affect different organs (liver, kidneys, skin, nervous system, etc.) (Chorus and Bartram, 1999). The harmful consequences of cyanobacterial toxins have been reported not only in laboratory bioassays but also in clinical and epidemiological studies that confirm the effects of toxic cyanobacteria on humans (Svircev et al., 2013). The production of certain cyanobacterial toxins is closely related to the amount of cyanobacterial biomass, especially when blooms appear (Spoerke and Rumack, 1985), which is why some organisations, such as the World Health Organization (WHO), have established criteria specifying the cyanobacteria abundance (cell number) and biovolume corresponding to different levels of threat to human health (Chorus and Bartram, 1999). A prompt response to dangerous blooms requires the accumulation of cyanobacterial biomass in water reservoirs to be continuously monitored. Numerous countries have developed regulations or guidelines to control cyanobacterial levels in water reservoirs, lakes, rivers and bathing zones by quantifying cyanobacterial biovolume, cell number,
11
and chlorophyll-a concentration (Chl a) or cyanotoxin (i.e., microcystins) concentrations (Chorus, 2012). Cyanobacterial development is greatly stimulated by eutrophication and nutrient excess in water (Vasconcelos, 2006). In limnological studies, the eutrophication level is assessed by quantifying the phytoplankton biomass (Willen, 1997) and Chl a (Hart, 1984). The Water Framework Directive (Directive 2000/60/EC[1]) and official regulations in many countries (Carvalho et al., 2013b) have specified that ecological status based on phytoplankton should be defined by measuring the biomass, composition and blooming events of the phytoplanktonic community. The Chl a concentration is a good parameter for evaluating the biomass of the entire phytoplanktonic community, but it does not provide information regarding the phytoplankton community composition, which must be assessed by biovolume estimations. Recent studies have shown that cyanobacterial biovolume can be used as a suitable metric for the assessment of the ecological status of lakes and reservoirs (Carvalho et al., 2013b). Planktonic cyanobacteria develop mainly in lentic water bodies, including lakes and water reservoirs, and in some cases, the toxin-producing cyanobacterial population can dominate the phytoplankton, which might affect humans that consume the water or make use of it for recreational purposes or crop irrigation.
Fig. 1 e Map of the Spanish watersheds and water reservoirs that passed the inclusion criteria. (1) Arcos; (2) Castro de las n Alto; (5) El Burguillo; (6) Santa Teresa; (7) Cíjara; (8) La Colada; (9) Alange; (10) Vega del Cogotas; (3) Cuerda del Pozo; (4) Ponto n; (11) Huesna; (12) Maria Cristina; (13) Bellus; (14) Borbollo n; (15) Guajaraz; (16) Navalca n; (17) Portaje; (18) Rivera de Jabalo ~ as; (22) Valmayor; (23) Cazalegas. Gata; (19) Rosarito; (20) Salor; (21) Valdecan
12
w a t e r r e s e a r c h 6 3 ( 2 0 1 4 ) 1 0 e2 0
C-phycocyanin, a photosynthetic pigment mostly produced by cyanobacteria, is the most representative cyanobacterial compound. Other types of phycocyanins can be found in rhodophyta species (R-phycocyanin), but they are easily differentiated from C-phycocyanin because they have different absorption peaks (Jiang et al., 2001). There are also trace concentrations of C-phycocyanin in some species of cryptophyceae (Rowan, 1989). In previously published works, field sampling methodologies were used to show that the Chl a concentration and phycocyanin concentration (PC) were closely related to the cyanobacterial biovolume and cell count (Brient et al., 2008). From March 2002 until April 2012, the Envisat satellite captured images of the entire world with the on board MERIS sensor, which was intended to monitor oceanic and coastal waters (Rast et al., 1999; Official ESA Website[2]). The MERIS sensor's narrow bandwidth permitted an accurate estimation of photosynthetic pigments in inland waters, although its spatial resolution (300 m) is not suitable for certain freshwater studies. In the absence of cloud cover, the Envisat MERIS sensor offers an advantage of gathering images of a large region, such as the Iberian Peninsula, every three days. This high frequency permits the collection of an extensive record of images for the analysis of phytoplanktonic community variation in water bodies located in regions with scarce cloud cover days, such as the countries along the Mediterranean coast. With the development of accurate algorithms, successful estimates of PC with MERIS were achieved compared to the PC results from field samples (Simis et al., 2007; Ruiz-Verdu´ et al., 2008; Guanter et al., 2010; Domínguez et al., 2011). The MERIS sensor produced a large collection of images that can recreate a time series of the cyanobacterial community seasonal distribution in water using either Chl a (Binding et al., 2011) or PC (Agha et al., 2012). Although a good correlation between PC and cyanobacterial cell count or biovolume has been found by sampling and measuring (Brient et al., 2008; Randolph et al., 2008), few researchers have directly compared the PC values measured by remote sensing to cell abundances or cyanobacterial biovolume, which are more useful variables for water quality monitoring (Willen, 1997; Chorus, 2012; Carvalho et al., 2013a). Although previous studies have compared the PC values estimated by airborne remote sensing to cyanobacterial cell counts with good results (Hunter et al., 2010), airborne remote sensing would be too expensive for
monitoring purposes, especially in countries with a large number of inland water bodies. Algorithms have also been developed for the MERIS sensor that relate Chl a levels to cyanobacteria biovolume (Matthews et al., 2012). However, until now, no studies have been performed that relate the PC values estimated by the MERIS sensor to cyanobacterial biovolume or cell counts. The few studies of inland waters that compared PC values with cyanobacterial cell numbers or compared Chl a with cyanobacterial biovolume only covered four or fewer water bodies (Hunter et al., 2010; Matthews et al., 2012). In this article, we use data from a large number of water bodies with different trophic statuses and different chemical characteristics distributed over a wide geographical scale. In Spain, there are more than 1500 water reservoirs, and cyanobacteria are frequently dominant in some of the reservoirs, especially in the southwest region of the Iberian Peninsula (De Hoyos et al., 2004; Quesada et al., 2004). Spanish water reservoirs have also been analysed by remote sensing for water quality purposes and to analyse cyanobacterial parameters, such as cyanobacterial biovolume or cell count (Ruiz-Verdu´ et al., 2008; Agha et al., 2012). In this work, water reservoirs throughout Spain from 2004 to 2010 were sampled, and a large database of field samples was produced that can be compared to the MERIS images. The objective of this research is to determine the relationship between cyanobacterial parameters obtained by both methods and to investigate whether this relationship depends on the taxonomic classification of the dominant biomass. This relationship can be used to estimate the cyanobacterial biovolume of a water reservoir from the images taken by the MERIS sensor.
2.
Methodology
2.1.
Available data
Data were obtained from two independent studies: a limnological survey in which numerous water reservoirs throughout Spain were sampled several times between 2004 and 2010 and another study in which all of the available MERIS sensor images from the Iberian Peninsula that were taken close to the time of the limnological research were retrieved.
Table 1 e Minimum, maximum, mean, median and number of samples (n) of the physical and biological parameters of the studied reservoirs. Water reservoir physical parameters Dam height (m) Surface (ha) Volume (hm3) Biological parameters Phycocyanin (mg/l) Chlorophyll-a (Laboratory Analysis) (mg/l) Chlorophyll-a (Remote Sensing) (mg/l) PC:Chl-a (MERIS) Cyanobacterial cell count (number of cells l/ml) Total cyanobacterial biovolume (mm3/l) Cyanobacterial percentage of total phytoplankton
Minimum 16 88 7.8
Maximum 98 7300 1670
Mean 49.93 1462.89 250.87
Median 47 692 58.65
n 23 23 23
Minimum 17.1 0.9 3.43 0.28 4716 0.2 3.09
Maximum 456.51 290.82 315.73 19.61 9933031 49.91 99.529
Mean 87.4 56.51 114.52 2.58 1018974 9.13 61.6
Median 69.64 25.95 23.31 2.34 142516 4.67 66.57
n 65 65 65 65 65 65 65
w a t e r r e s e a r c h 6 3 ( 2 0 1 4 ) 1 0 e2 0
13
14
2.2.
w a t e r r e s e a r c h 6 3 ( 2 0 1 4 ) 1 0 e2 0
Field sampling and analysis methodology
Field sampling occurred during the summer months (usually from June to September) over one or more years in numerous water reservoirs. The sampling points were located 100 m from the dam, and samples were collected from a single point located at the maximum Chl a concentration depth (as defined by the profile obtained with the probe YSI6600 V2). Sampling points were always within the euphotic zone (defined as 2.5 times the Secchi disc depth). No scum episodes were detected during the samplings. Phytoplankton was counted in the laboratory according to € hl sedimentation method (Sournia, 1978), and the Utermo phytoplanktonic organisms were identified at the finest taxonomic level possible. Biovolume was calculated by producing an approximate simple or compound geometrical shape for each species (Hillebrand et al., 1999). Chl a was extracted with 90% acetone and quantified with a spectrophotometer using the trichromatic equations (Parsons and Strickland, 1968).
2.3.
MERIS image methodology
The MERIS images used in this report are level 1b full resolution products (FR L1b). Suitable FR L1b images from the sampled water reservoirs were acquired from Earthnet OnLine Interactive (EOLi). Atmospheric correction (SCAPEM_B2), geometrical correction (Georeferencing) and mosaicking methodologies were applied to the images. SCAPE-M_B2 is an improved version of SCAPE-M that has a correction in band 2 (Domínguez et al., 2011) (see below). SCAPE-M was developed and validated in several European water bodies (Guanter et al., 2010), and the SCAPE-M equation is as follows: LTOA ¼ L0 þ
1 rs Edir mil þ Edif T[ p 1 Srs
(1)
where LTOA represents the surface reflectance images derived from the top-of atmosphere radiance; L0 is the atmospheric path radiance; mil is the cosine of the illumination zenith angle qil measured between the solar ray and surface normal; Edirmil and Edif are the direct and diffuse fluxes, respectively, arriving at the surface; S is the atmospheric spherical albedo, which represents the reflectance of the atmosphere for isotropic light entering from the surface; T[ is the total atmospheric transmittance (for diffuse plus direct radiation) in the observation direction; and rs is the surface reflectance. SCAPE-M was corrected for the MERIS band 2 by means of an interpolation between the values of band 1 and band 3; if the original value is used, maximum values for vegetation and water are obtained because those values correspond to the absorption of Chl a. Thus, the correction with SCAPE-M is performed twice, and the assigned value of band 2
corresponds to the interpolated value of band 2 minus the value obtained without interpolation (Domínguez et al., 2011). The result of this improvement is the atmospheric correction algorithm SCAPE-M_B2. A water mask was used in the images to separate water zones from land areas, including the dam. Water/land pixels were discriminated with a simple threshold in the nearinfrared band (Agha et al., 2012). From each water pixels, the Chl a and PC values were estimated with the following algorithms (Domínguez et al., 2011) to generate MERIS thematic mappings: when Chl a 19:34 mg m3 : Chl a ¼ 19:34e6:1257
(2)
when Chl a 19:34 mg m3 : Chl a ¼ 19:34e5:2044½ðB9B7Þ=ðB9þB7Þ
(3)
PC ¼ 46: 478e5:186½ðB9B6Þ=ðB9þB6Þ
(4)
where B2 is the fluorescence of the first absorption peak of chlorophyll-a at approximately 442 nm (MERIS band 2); B5 is the maximum solar radiation that penetrates water at 560 nm and corresponds to the green band (MERIS band 5); B6 corresponds to the absorption peak of PC centred at 620 nm (MERIS band 6); B7 is the second absorption peak of chlorophyll-a at approximately 665 nm (MERIS band 7); and B9 is the fluorescence of chlorophyll-a at approximately 705 nm (MERIS band 9). The Chl a algorithms were calibrated with HPLC field data, and the PC algorithm was calibrated with a calibrating fluorometer (Domínguez et al., 2011). The Chl a and PC values retrieved from the MERIS thematic mapping correspond to the water pixels located at a midpoint near the dam, and they cover the points where field samples were obtained. The MERIS sensor captured the first optical thickness, which is equivalent to 0.6 times the Secchi disc depth.
2.4.
Data selection and area of study
Within the database, two conditions for image inclusion were established. 1) The MERIS image must have a clear pixel that covers the field sampling point. Consequently, water reservoirs with narrow dams or small surfaces were rejected. 2) Only field sampling data with cyanobacterial biovolume greater than 0.2 mm3/l are used. This threshold was selected because it corresponds to the first alert level for a cyanobacterial hazard. Using these conditions, we selected 23 different Spanish water reservoirs and 65 field samples (Fig. 1, Table 1). Twelve of the water reservoirs are in the Tajo river watershed, and 21 are in the west-central region of Spain, which is characterised by a hot-summer Mediterranean climate. Regions with this
Fig. 2 e Comparisons of chlorophyll-a concentrations between both methodologies. Both linear regressions were forced to begin at the origin (0,0) for validation purposes. The broken red line represents y ¼ x (RMSE ¼ root mean square error). a) Comparison of chlorophyll-a concentration measuring methodologies. n ¼ 52; p < 0.001; RMSE ¼ 0.024. b) Comparison of normalised chlorophyll-a concentration measuring methodologies. n ¼ 52; p < 0.001; RMSE ¼ 0.031. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
15
w a t e r r e s e a r c h 6 3 ( 2 0 1 4 ) 1 0 e2 0
Table 2 e Summary of the time differences between both methodologies (n ¼ 65).
Days of difference between the MERIS image and field sampling (absolute) Matching data information
Minimum
Maximum
Mean
Median
0
14
3.27
3
Matching sample dates
Satellite image retrieval not blocked by cloud cover (not including matching sample dates) 24
Total times that satellite image retrieval was blocked by cloud cover
Total times that satellite image retrieval was blocked by cloud cover more than once 6
8
43
it is the reservoir with the densest data as a result of having the most consecutive summer field samples available in the database (from 2007 to 2010).
climate typically experience hot, dry summers and mild, wet winters, and precipitation is higher during the colder months. Comparing the MERIS sensor images to the field data is especially useful in regions with climates characterised by infrequent cloud cover days because additional MERIS images would be available for the field sampling date.
3.
Results and discussion
2.5.
3.1.
Chlorophyll-a comparison
Results validation and statistical analysis
We validated the empirical results in the complete dataset using an external dataset from a hypertrophic water reservoir, the Rosarito reservoir. The Rosarito reservoir belongs to the Tietar River and is located in the Tajo watershed; the dam height is 38 m, and it has a water surface of 1475 km2 and a capacity of 92 hm3. The Rosarito reservoir was chosen because
Lab-measured and remote-sensing estimated Chl a concentrations (Fig. 2a) were compared, and a strong correlation between them was found (R2 ¼ 0.967; p < 0.001). Even when data are normalised, the correlation is still strong (R2 ¼ 0.947; p < 0.001) (Fig. 2b). Therefore, we can conclude that outliers do not produce strong leverage against the R2 value.
Fig. 3 e Correlation between phycocyanin concentration estimated by remote sensing and total cyanobacterial biovolume. n ¼ 65; p < 0.001; RMSE ¼ 0.06. (RMSE ¼ root mean square error).
16
w a t e r r e s e a r c h 6 3 ( 2 0 1 4 ) 1 0 e2 0
It is well known that the distribution of cyanobacteria in a water body is not perfectly homogeneous, and papers have reported that a patchy distribution of scum can produce unreliable pigment estimations by the MERIS sensor (Kutser, 2004). However, in the present work, no scum episodes were detected during the sampling periods. Therefore, no impact on the correlation was expected as a result of the heterogeneous distribution of scum. Another possible cause of error is the vertical distribution of the cyanobacterial community. Field samples were collected from the maximum Chl a layer within the euphotic zone, which has a maximum depth equal to 2.5 times the depth of the Secchi disk. However, the MERIS sensor covers the first optical thickness, which reaches 0.6 times the depth of the Secchi disk. Therefore, when most of the cyanobacterial community is located deeper than the first optical thickness, the community might not be detected by the MERIS sensor. However, when heterogeneity in the vertical cyanobacterial distribution is not intense, recent publications have found a good agreement between the MERIS sensor pigment estimation and field samples (Ruiz-Verdu´ et al., 2008; Guanter et al., 2010; Domínguez et al., 2011; Agha et al., 2012; Dekker and Hestir, 2012; Matthews et al., 2012; Mishra et al., 2013; Ali et al., 2014). In this work, field samples were collected at a certain distance from the dams where the distribution of the phytoplanktonic communities is generally expected to be more
homogenous (Kimmel et al., 1990). Moreover, the algorithms used to estimate both the Chl a and PC values in our work have been proven reliable and accurate compared to field samples (Domínguez et al., 2011). Major differences between the Chl a levels in the field and MERIS data were considered to be a result of heterogeneity in the cyanobacterial community, both at the temporal (Table 2) and spatial scales, and no relevant changes in the phytoplankton composition were expected given the short time lag between both measurements. Our data show a high correlation between the Chl a datasets (Fig. 2), and the MERIS data can be considered highly reflective of the field-sampled community.
3.2. Phycocyanin and cyanobacterial biovolume correlation The PC values estimated by remote sensing and the total cyanobacterial biovolume measured in the field samples (Fig. 3) were found to be significantly correlated (R2 ¼ 0.6219; p < 0.001), and the values are related by the following function: Cyanobacterial biovolume ¼ ð132:39$PCÞ 1825:1
(5)
Previous reports also found high correlations between the cyanobacterial biovolume and PC values estimated by waterborne radiometry analysis (Randolph et al., 2008).
Fig. 4 e Correlation between phycocyanin concentrations estimated by remote sensing and number of cells per millilitre. n ¼ 65; p < 0.001; RMSE ¼ 0.19. (RMSE ¼ root mean square error).
w a t e r r e s e a r c h 6 3 ( 2 0 1 4 ) 1 0 e2 0
The potential regression between the number of cells per volume and PC showed the strongest determination coefficient. However, it still shows a weak correlation(Fig. 4), although it was significant(R2 ¼ 0.362; p < 0.001). This relationship was slightly weaker than that found in other studies comparing the PC values estimated by remote sensing with cell counts (Hunter et al., 2010), and it is remarkably weaker than other reported field studies that used probe samples (Brient et al., 2008). However, this may have been a result of our study including 23 different water bodies with a heterogeneous species composition, which may scatter the data distribution. The phycobiliprotein cell quota can vary drastically for different cyanobacteria (Patel et al., 2005). Therefore, our results clearly identify biovolume as the most reliable biomass parameter to be estimated by remote sensing techniques in a diverse group of water bodies. When we considered the taxonomic distribution of the cyanobacteria found in the water bodies and subdivided the dataset by cyanobacterial order (Fig. 5), we found the best correlation between biovolume and PC for the order Oscillatoriales (R2 ¼ 0.726, p < 0.001). Nostocales and Chroococcales showed weaker correlations (Nostocales: R2 ¼ 0.422, p < 0.001; Chroococcales: R2 ¼ 0.517; p < 0.05). Although the correlation of Nostocales is the weakest of the three orders, the pattern described
17
by the linear regression and biovolume/pigment ratio is similar to that of Oscillatoriales and total cyanobacteria. The weaker correlation of Chroococcales and the different pattern followed by its biovolume/pigment correlation could be caused by the dominance of Microcystis aeruginosa (which was dominant in 6 of the 12 Chroococcales samples). Microcystis aeruginosa has been described as highly variable in terms of its cellular pigment content, even when grown in laboratory conditions (BanaresEspana et al., 2007). It is also important to note that Microcystis aeruginosa forms large 3D colonies that can exhibit unexpected optical behaviour (scattering within the colony) and can make cell counting difficult. The relatively low number of samples from water bodies dominated by Chroococcales (Chroococcales ¼ 12; Nostocales ¼ 30; Oscillatoriales ¼ 22) could also partly explain the weaker correlation. Considering the statistical robustness of our results, we can conclude that the equation given by the linear regression between the cyanobacterial biovolume and PC values estimated by remote sensing is a good method for calculating cyanobacterial biovolume using remote sensing tools. The dataset used in our study is diverse in terms of cyanobacterial abundance and taxonomic distribution, so our equation should be robust for assessing cyanobacterial biomass (as biovolume) in a wide range of ecosystems. This result represents an improvement
Fig. 5 e Correlation between phycocyanin concentrations and total cyanobacterial biovolume depending on the dominant order (RMSE ¼ root mean square error). (Nostocales: n ¼ 30, p < 0.001, RMSE ¼ 0.08; Oscillatoriales: n ¼ 30, p < 0.001, RMSE ¼ 0.07; Chroococcales: n ¼ 13, p < 0.05, RMSE ¼ 0.08).
18
w a t e r r e s e a r c h 6 3 ( 2 0 1 4 ) 1 0 e2 0
Fig. 6 e Time line comparing the estimated cyanobacterial biovolume (calculated from phycocyanin estimated by MERIS) and field sampled cyanobacterial biovolume.
over previously published methods, which were based on a limited number of water bodies and community types (Randolph et al., 2008; Hunter et al., 2010; Matthews et al., 2012).
3.3.
Validation
The equation obtained with the entire dataset was applied to the MERIS imagery of the Rosarito reservoir to obtain an
estimated cyanobacterial biovolume time series. This time series was compared to the biovolume data measured in an extensive sampling schedule spanning 4 years. The results (Fig. 6) show a high correlation between the biovolume measured in the field samples and the biovolume estimated by applying the equation to the satellite imagery. However, some discrepancies were found in 2010 that were most likely a result of a significant difference between the sampling dates and
Fig. 7 e Blue line (left y axis): determination coefficients (R2) of the correlation between the estimated cyanobacterial biovolume and field sampled cyanobacterial biovolume in the Rosarito reservoir depending on the days of difference between both date. Red broken line (right y axis): number of samples for each correlation. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
w a t e r r e s e a r c h 6 3 ( 2 0 1 4 ) 1 0 e2 0
image acquisition dates. In fact, when data are disaggregated by the time lag, the determination coefficients of the respective correlations between the field sampling and imaging show a clear decrease after a difference of 9 days between the acquisition of the data with both methodologies (Fig. 7). It is obvious that when sampling and imaging occur on the same day, the correlation is very good (R2 ¼ 0.72; p < 0.001). When all of the time lags are considered, the determination coefficients are weaker, but they are still significant until a difference of 9 days (R2 varies between 0.51 and 0.55; p < 0.001). However, when the time lag spans longer than 9 days, there is no significant correlation (R2 ¼ 0.1676; p ¼ 0.06). Although the MERIS sensor is now inoperative, Sentinel-3 is expected to have a similar sensor available that will also retrieve L1b products after 1 h of scanning (Donlon et al., 2012). Therefore, the Sentinel-3 mission might be a useful monitoring tool in support of management strategies for water bodies larger than 9 ha, especially for countries with few cloudy days. The previously recorded MERIS images can also represent a useful data source for understanding cyanobacterial dynamics in certain water bodies over long-term periods; in certain cases, up to 10 years of high frequency data can become crucial to the development of management strategies and increased forecasting capabilities.
4.
Conclusions
In this work, we define and validate the use of MERIS imagery to estimate the cyanobacterial biovolume in freshwater ecosystems under non-scum conditions. This approach is based on the detection of the cyanobacterial pigment PC by means of the MERIS sensor, and we have adapted this tool to calculate cyanobacterial biovolume, an important parameter for limnological and water quality studies. In our analysis, the correlation between the measured cyanobacterial biovolume and PC values estimated from the MERIS imagery was found to be a robust proxy for cyanobacterial biomass estimation. The regression equation was built by using data from 23 aquatic ecosystems in the Iberian Peninsula that were dominated by diverse cyanobacterial communities. A validation exercise conducted with the time series data from an external water reservoir indicated that the equation was useful for estimating cyanobacterial biomass and that it had excellent results when the time lag between the field measurements and MERIS imaging was less than 9 days. Considering that our data covered a wide area of different lithological and trophic characteristics, included 89 different cyanobacteria species (of which 23 different species were dominant) and produced positive results from the validation test, we believe that these results could be transferable to other water reservoirs.
Acknowledgements This research was funded by a grant from the Centre for Studies and Experimentation of Public Works (CEDEX) of the Urbanism Ministry of Spain. Special thanks to the Centre for
19
Hydrographic Studies and everyone who helped to obtain the data from the water reservoirs for this research.
Appendix A. Supplementary data Supplementary data related to this article can be found at http://dx.doi.org/10.1016/j.watres.2014.06.001.
references
s, S., Wo € rmer, L., Domínguez, J.A., Quesada, A., 2012. Agha, R., Cire Multi-scale strategies for the monitoring of freshwater cyanobacteria: reducing the sources of uncertainty. Water Res. 46 (9), 3043e3053. Ali, K., Witter, D., Ortiz, J., 2014. Application of empirical and semi-analytical algorithms to MERIS data for estimating chlorophyll a in case 2 waters of Lake Erie. Environ. Earth Sci. 71 (9), 4209e4220. Banares-Espana, E., Lopez-Rodas, V., Costas, E., Salgado, C., Flores-Moya, A., 2007. Genetic variability associated with photosynthetic pigment concentration, and photochemical and nonphotochemical quenching, in strains of the cyanobacterium Microcystis aeruginosa. FEMS Microbiol. Ecol. 60 (3), 449e455. Binding, C.E., Greenberg, T.A., Bukata, R.P., 2011. Time series analysis of algal blooms in lake of the woods using the MERIS maximum chlorophyll index. J. Plankton Res. 33 (12), 1847e1852. Brient, L., Lengronne, M., Bertrand, E., Rolland, D., Sipel, A., Steinmann, D., Baudin, I., Legeas, M., Le Rouzic, B., Bormans, M., 2008. A phycocyanin probe as a tool for monitoring cyanobacteria in freshwater bodies. J. Environ. Monit. 10 (2), 248e255. Carvalho, L., McDonald, C., de Hoyos, C., Mischke, U., Phillips, G., Borics, G., Poikane, S., Skjelbred, B., Solheim, A.L., Van Wichelen, J., Cardoso, A.C., 2013a. Sustaining recreational quality of European lakes: minimizing the health risks from algal blooms through phosphorus control. J. Appl. Ecol. 50 (2), 315e323. Carvalho, L., Poikane, S., Solheim, A.L., Phillips, G., Borics, G., Catalan, J., De Hoyos, C., Drakare, S., Dudley, B.J., Jarvinen, M., Laplace-Treyture, C., Maileht, K., McDonald, C., Mischke, U., Moe, J., Morabito, G., Noges, P., Noges, T., Ott, I., Pasztaleniec, A., Skjelbred, B., Thackeray, S.J., 2013b. Strength and uncertainty of phytoplankton metrics for assessing eutrophication impacts in lakes. Hydrobiologia 704 (1), 127e140. Chorus, I., 2012. Current Approaches to Cyanotoxin Risk Assessment, Risk Management and Regulations in Different Countries. Federal Environment Agency, Berlin. Chorus, I., Bartram, J., 1999. Toxic Cyanobacteria in Water. Spon Press. De Hoyos, C., N., A.I., Aldasoro, J.J., 2004. Cyanobacteria distribution and abundance in the Spanish water reservoirs during thermal stratification. Limnetica 23 (1e2), 119e132. Dekker, A.G., Hestir, E.L., 2012. Evaluating the Feasibility of Systematic Inland Water Quality Monitoring with Satellite Remote Sensing. Domínguez, J.A., Alonso, C., Alonso, A., 2011. Remote sensing as a tool for monitoring water quality parameters for Mediterranean Lakes of European Union water framework directive (WFD) and as a system of surveillance of cyanobacterial harmful algae blooms (SCyanoHABs). Environ. Monit. Assess. 181 (1e4), 317e334.
20
w a t e r r e s e a r c h 6 3 ( 2 0 1 4 ) 1 0 e2 0
Donlon, C., Berruti, B., Buongiorno, A., Ferreira, M.H., Femenias, P., Frerick, J., Goryl, P., Klein, U., Laur, H., Mavrocordatos, C., Nieke, J., Rebhan, H., Seitz, B., Stroede, J., Sciarra, R., 2012. The global monitoring for environment and security (GMES) sentinel-3 mission. Remote Sens. Environ. 120, 37e57. Guanter, L., Ruiz-Verdu, A., Odermatt, D., Giardino, C., Simis, S., Estelles, V., Heege, T., Dominguez-Gomez, J.A., Moreno, J., 2010. Atmospheric correction of ENVISAT/MERIS data over inland waters: validation for European lakes. Remote Sens. Environ. 114 (3), 467e480. Hart, B.T., 1984. Eutrophication of waters e monitoring, assessment and control e OECD. Search 15 (1e2), 50. Hillebrand, H., Du¨rselen, C.-D., Kirschtel, D., Pollingher, U., Zohary, T., 1999. Biovolume calculation for pelagic and benthic microalgae. J. Phycol. 35 (2), 403e424. Hunter, P.D., Tyler, A.N., Carvalho, L., Codd, G.A., Maberly, S.C., 2010. Hyperspectral remote sensing of cyanobacterial pigments as indicators for cell populations and toxins in eutrophic lakes. Remote Sens. Environ. 114, 2705e2718. Jiang, T., Zhang, J.P., Chang, W.R., Liang, D.C., 2001. Crystal structure of R-phycocyanin and possible energy transfer pathways in the phycobilisome. Biophys. J. 81 (2), 1171e1179. Kimmel, B.L., Lind, O.T., Paulson, L.J., 1990. Reservoir primary production. In: Thorton, K.W., Kimmel, B.L., Payne, F.E. (Eds.), Reservoir Limnology: Ecological Perspectives. John Wiley & Sons, Inc, pp. 133e193. Kutser, T., 2004. Quantitative detection of chlorophyll in cyanobacterial blooms by satellite remote sensing. Limnol. Oceanogr. 49 (6), 2179e2189. Matthews, M.W., Bernard, S., Robertson, L., 2012. An algorithm for detecting trophic status (chlorophyll-a), cyanobacterialdominance, surface scums and floating vegetation in inland and coastal waters. Remote Sens. Environ. 124, 637e652. Mishra, S., Mishra, D.R., Lee, Z., Tucker, C.S., 2013. Quantifying cyanobacterial phycocyanin concentration in turbid productive waters: a quasi-analytical approach. Remote Sens. Environ. 133, 141e151. Parsons, T.R., Strickland, J.D.H., 1968. A Practical Handbook of Seawater Analysis. Pigment analysis. Otawa. Supply and Services Canada. Patel, A., Mishra, S., Pawar, R., Ghosh, P.K., 2005. Purification and characterization of C-phycocyanin from cyanobacterial species of marine and freshwater habitat. Protein Expr. Purif. 40 (2), 248e255.
Quesada, A., Sanchis, D., Carrasco, D., 2004. Cyanobacteria in Spanish reservoirs. How frequently are toxic? Limnetica 23 (1e2), 109e118. Randolph, K., Wilson, J., Tedesco, L., Li, L., Pascual, D.L., Soyeux, E., 2008. Hyperspectral remote sensing of cyanobacteria in turbid productive water using optically active pigments, chlorophyll a and phycocyanin. Remote Sens. Environ. 112, 4009e4019. Rast, M., Bezy, J.L., Bruzzi, S., 1999. The ESA medium resolution imaging spectrometer MERIS e a review of the instrument and its mission. Int. J. Remote Sens. 20 (9), 1681e1702. Rowan, K.S., 1989. Photosynthetics Pigments of Algae. Cambridge University Press, Cambridge. ~ aRuiz-Verdu´, A., Simis, S.G.H., de Hoyos, C., Gons, H.J., Pen Martínez, R., 2008. An evaluation of algorithms for the remote sensing of cyanobacterial biomass. Remote Sens. Environ. 112 (11), 3996e4008. Simis, S.G.H., Ruiz-Verdu, A., Dominguez-Gomez, J.A., PenaMartinez, R., Peters, S.W.M., Gons, H.J., 2007. Influence of phytoplankton pigment composition on remote sensing of cyanobacterial biomass. Remote Sens. Environ. 106 (4), 414e427. Sournia, A., 1978. Phytoplankton Manual, vol. 1. Paris. Spoerke, D.G., Rumack, B.H., 1985. Blue-green algae poisoning. J. Emerg. Med. 2 (5), 353e355. Svircev, Z., Simeunovic, J., Subakov-Simic, G., Krstic, S., Pantelic, D., Dulic, T., 2013. Cyanobacterial blooms and their toxicity in Vojvodina lakes, Serbia. Int. J. Environ. Res. 7 (3), 845e858. Vasconcelos, V., 2006. Eutrophication, toxic cyanobacteria and cyanotoxins: when ecosystems cry for help. Limnetica 25 (1e2), 425e432. Whitton, B.A., Potts, M., 2002. Their Diversity in Time and Space, vol. 1. Springer, Netherlands. Willen, E., 1997. Phytoplankton in water quality assessment, an indicator concept. Phycologia 36 (4), 123e124.
url's
http://eur-lex.europa.eu/LexUriServ/LexUriServ.do?uri¼CELEX: 32000L0060:en:HTML. https://earth.esa.int/web/guest/missions/esa-operational-eomissions/envisat/instruments/meris/design. https://earth.esa.int/web/guest/missions/esa-future-missions/ sentinel-3.
Available online at www.sciencedirect.com
ScienceDirect journal homepage: www.elsevier.com/locate/watres
Estimation of cyanobacteria biovolume in water reservoirs by MERIS sensor M. Medina-Cobo a, J.A. Domínguez b, A. Quesada c, C. de Hoyos a,* ficos (CEDEX), Paseo bajo de la Virgen del Puerto 3, 28005 Madrid, Spain Centro de Estudios Hidrogra Mathematical and Fluid Physics Department, Universidad Nacional de Educacion a Distancia (UNED), Senda del Rey Street 9, 28040 Madrid, Spain c Dpt. Biology, Universidad Autonoma de Madrid, Calle Darwin 2, 28049 Madrid, Spain a
b
article info
abstract
Article history:
Planktonic cyanobacteria primarily develop in lentic water bodies, such as lakes and water
Received 22 January 2014
reservoirs. In certain instances, toxin-producing cyanobacterial populations might domi-
Received in revised form
nate the phytoplankton community. Satellite remote sensing is a useful tool for large
29 May 2014
spatial scale monitoring of cyanobacteria, and the MERIS sensor from the Envisat satellite
Accepted 2 June 2014
has taken worldwide images at a high frequency for over 10 years. This short time lapse
Available online 11 June 2014
image collection has provided an extensive record of images for the analysis of variation in the cyanobacterial communities in water reservoirs for management and scientific pur-
Keywords:
poses. The objective of this work is to determine the relationship between measured
Cyanobacteria
cyanobacterial biomass as biovolume and the estimations derived from MERIS imagery.
Biovolume
This study encompasses two independent studies relying on data from 23 water reservoirs.
Phycocyanin
First, a long-term global limnological research study was conducted that provided a field
Water reservoirs
data collection that included cyanobacterial biovolume, among other variables. Second, a
MERIS sensor
survey was conducted that applied the processed images derived from the Envisat MERIS
Time series
sensor. The chlorophyll-a (Chl a) content and phycocyanin concentration (PC) were estimated from the MERIS images. The PC estimated by remote sensing and total cyanobacterial biovolume measured from the field samples were found to be significantly correlated (R2 ¼ 0.6219; p < 0.001). No relevant differences were found among the taxonomical groups, which indicated that this tool provides accurate estimations irrespective of the cyanobacterial group. For validation, the algorithm derived from the entire dataset was applied to the MERIS image dataset of the Rosarito reservoir. An estimated cyanobacterial biovolume time series was performed and compared to the biovolume data collected in an extensive sampling schedule spanning 4 years. The results indicated a strong correlation (R2 ¼ 0.72; p < 0.001) between the measured and estimated data acquired on the same day. © 2014 Elsevier Ltd. All rights reserved.
* Corresponding author. E-mail addresses: [email protected] (M. Medina-Cobo), [email protected] (J.A. Domínguez), antonio.quesada@ uam.es (A. Quesada), [email protected] (C. de Hoyos). http://dx.doi.org/10.1016/j.watres.2014.06.001 0043-1354/© 2014 Elsevier Ltd. All rights reserved.
w a t e r r e s e a r c h 6 3 ( 2 0 1 4 ) 1 0 e2 0
1.
Introduction
Cyanobacteria are a phylum of photosynthetic prokaryotic organisms distributed worldwide (Whitton and Potts, 2002). In terms of human health risk assessment, cyanobacteria can pose a hazard when they produce cyanotoxins, a family of secondary metabolites that can affect different organs (liver, kidneys, skin, nervous system, etc.) (Chorus and Bartram, 1999). The harmful consequences of cyanobacterial toxins have been reported not only in laboratory bioassays but also in clinical and epidemiological studies that confirm the effects of toxic cyanobacteria on humans (Svircev et al., 2013). The production of certain cyanobacterial toxins is closely related to the amount of cyanobacterial biomass, especially when blooms appear (Spoerke and Rumack, 1985), which is why some organisations, such as the World Health Organization (WHO), have established criteria specifying the cyanobacteria abundance (cell number) and biovolume corresponding to different levels of threat to human health (Chorus and Bartram, 1999). A prompt response to dangerous blooms requires the accumulation of cyanobacterial biomass in water reservoirs to be continuously monitored. Numerous countries have developed regulations or guidelines to control cyanobacterial levels in water reservoirs, lakes, rivers and bathing zones by quantifying cyanobacterial biovolume, cell number,
11
and chlorophyll-a concentration (Chl a) or cyanotoxin (i.e., microcystins) concentrations (Chorus, 2012). Cyanobacterial development is greatly stimulated by eutrophication and nutrient excess in water (Vasconcelos, 2006). In limnological studies, the eutrophication level is assessed by quantifying the phytoplankton biomass (Willen, 1997) and Chl a (Hart, 1984). The Water Framework Directive (Directive 2000/60/EC[1]) and official regulations in many countries (Carvalho et al., 2013b) have specified that ecological status based on phytoplankton should be defined by measuring the biomass, composition and blooming events of the phytoplanktonic community. The Chl a concentration is a good parameter for evaluating the biomass of the entire phytoplanktonic community, but it does not provide information regarding the phytoplankton community composition, which must be assessed by biovolume estimations. Recent studies have shown that cyanobacterial biovolume can be used as a suitable metric for the assessment of the ecological status of lakes and reservoirs (Carvalho et al., 2013b). Planktonic cyanobacteria develop mainly in lentic water bodies, including lakes and water reservoirs, and in some cases, the toxin-producing cyanobacterial population can dominate the phytoplankton, which might affect humans that consume the water or make use of it for recreational purposes or crop irrigation.
Fig. 1 e Map of the Spanish watersheds and water reservoirs that passed the inclusion criteria. (1) Arcos; (2) Castro de las n Alto; (5) El Burguillo; (6) Santa Teresa; (7) Cíjara; (8) La Colada; (9) Alange; (10) Vega del Cogotas; (3) Cuerda del Pozo; (4) Ponto n; (11) Huesna; (12) Maria Cristina; (13) Bellus; (14) Borbollo n; (15) Guajaraz; (16) Navalca n; (17) Portaje; (18) Rivera de Jabalo ~ as; (22) Valmayor; (23) Cazalegas. Gata; (19) Rosarito; (20) Salor; (21) Valdecan
12
w a t e r r e s e a r c h 6 3 ( 2 0 1 4 ) 1 0 e2 0
C-phycocyanin, a photosynthetic pigment mostly produced by cyanobacteria, is the most representative cyanobacterial compound. Other types of phycocyanins can be found in rhodophyta species (R-phycocyanin), but they are easily differentiated from C-phycocyanin because they have different absorption peaks (Jiang et al., 2001). There are also trace concentrations of C-phycocyanin in some species of cryptophyceae (Rowan, 1989). In previously published works, field sampling methodologies were used to show that the Chl a concentration and phycocyanin concentration (PC) were closely related to the cyanobacterial biovolume and cell count (Brient et al., 2008). From March 2002 until April 2012, the Envisat satellite captured images of the entire world with the on board MERIS sensor, which was intended to monitor oceanic and coastal waters (Rast et al., 1999; Official ESA Website[2]). The MERIS sensor's narrow bandwidth permitted an accurate estimation of photosynthetic pigments in inland waters, although its spatial resolution (300 m) is not suitable for certain freshwater studies. In the absence of cloud cover, the Envisat MERIS sensor offers an advantage of gathering images of a large region, such as the Iberian Peninsula, every three days. This high frequency permits the collection of an extensive record of images for the analysis of phytoplanktonic community variation in water bodies located in regions with scarce cloud cover days, such as the countries along the Mediterranean coast. With the development of accurate algorithms, successful estimates of PC with MERIS were achieved compared to the PC results from field samples (Simis et al., 2007; Ruiz-Verdu´ et al., 2008; Guanter et al., 2010; Domínguez et al., 2011). The MERIS sensor produced a large collection of images that can recreate a time series of the cyanobacterial community seasonal distribution in water using either Chl a (Binding et al., 2011) or PC (Agha et al., 2012). Although a good correlation between PC and cyanobacterial cell count or biovolume has been found by sampling and measuring (Brient et al., 2008; Randolph et al., 2008), few researchers have directly compared the PC values measured by remote sensing to cell abundances or cyanobacterial biovolume, which are more useful variables for water quality monitoring (Willen, 1997; Chorus, 2012; Carvalho et al., 2013a). Although previous studies have compared the PC values estimated by airborne remote sensing to cyanobacterial cell counts with good results (Hunter et al., 2010), airborne remote sensing would be too expensive for
monitoring purposes, especially in countries with a large number of inland water bodies. Algorithms have also been developed for the MERIS sensor that relate Chl a levels to cyanobacteria biovolume (Matthews et al., 2012). However, until now, no studies have been performed that relate the PC values estimated by the MERIS sensor to cyanobacterial biovolume or cell counts. The few studies of inland waters that compared PC values with cyanobacterial cell numbers or compared Chl a with cyanobacterial biovolume only covered four or fewer water bodies (Hunter et al., 2010; Matthews et al., 2012). In this article, we use data from a large number of water bodies with different trophic statuses and different chemical characteristics distributed over a wide geographical scale. In Spain, there are more than 1500 water reservoirs, and cyanobacteria are frequently dominant in some of the reservoirs, especially in the southwest region of the Iberian Peninsula (De Hoyos et al., 2004; Quesada et al., 2004). Spanish water reservoirs have also been analysed by remote sensing for water quality purposes and to analyse cyanobacterial parameters, such as cyanobacterial biovolume or cell count (Ruiz-Verdu´ et al., 2008; Agha et al., 2012). In this work, water reservoirs throughout Spain from 2004 to 2010 were sampled, and a large database of field samples was produced that can be compared to the MERIS images. The objective of this research is to determine the relationship between cyanobacterial parameters obtained by both methods and to investigate whether this relationship depends on the taxonomic classification of the dominant biomass. This relationship can be used to estimate the cyanobacterial biovolume of a water reservoir from the images taken by the MERIS sensor.
2.
Methodology
2.1.
Available data
Data were obtained from two independent studies: a limnological survey in which numerous water reservoirs throughout Spain were sampled several times between 2004 and 2010 and another study in which all of the available MERIS sensor images from the Iberian Peninsula that were taken close to the time of the limnological research were retrieved.
Table 1 e Minimum, maximum, mean, median and number of samples (n) of the physical and biological parameters of the studied reservoirs. Water reservoir physical parameters Dam height (m) Surface (ha) Volume (hm3) Biological parameters Phycocyanin (mg/l) Chlorophyll-a (Laboratory Analysis) (mg/l) Chlorophyll-a (Remote Sensing) (mg/l) PC:Chl-a (MERIS) Cyanobacterial cell count (number of cells l/ml) Total cyanobacterial biovolume (mm3/l) Cyanobacterial percentage of total phytoplankton
Minimum 16 88 7.8
Maximum 98 7300 1670
Mean 49.93 1462.89 250.87
Median 47 692 58.65
n 23 23 23
Minimum 17.1 0.9 3.43 0.28 4716 0.2 3.09
Maximum 456.51 290.82 315.73 19.61 9933031 49.91 99.529
Mean 87.4 56.51 114.52 2.58 1018974 9.13 61.6
Median 69.64 25.95 23.31 2.34 142516 4.67 66.57
n 65 65 65 65 65 65 65
w a t e r r e s e a r c h 6 3 ( 2 0 1 4 ) 1 0 e2 0
13
14
2.2.
w a t e r r e s e a r c h 6 3 ( 2 0 1 4 ) 1 0 e2 0
Field sampling and analysis methodology
Field sampling occurred during the summer months (usually from June to September) over one or more years in numerous water reservoirs. The sampling points were located 100 m from the dam, and samples were collected from a single point located at the maximum Chl a concentration depth (as defined by the profile obtained with the probe YSI6600 V2). Sampling points were always within the euphotic zone (defined as 2.5 times the Secchi disc depth). No scum episodes were detected during the samplings. Phytoplankton was counted in the laboratory according to € hl sedimentation method (Sournia, 1978), and the Utermo phytoplanktonic organisms were identified at the finest taxonomic level possible. Biovolume was calculated by producing an approximate simple or compound geometrical shape for each species (Hillebrand et al., 1999). Chl a was extracted with 90% acetone and quantified with a spectrophotometer using the trichromatic equations (Parsons and Strickland, 1968).
2.3.
MERIS image methodology
The MERIS images used in this report are level 1b full resolution products (FR L1b). Suitable FR L1b images from the sampled water reservoirs were acquired from Earthnet OnLine Interactive (EOLi). Atmospheric correction (SCAPEM_B2), geometrical correction (Georeferencing) and mosaicking methodologies were applied to the images. SCAPE-M_B2 is an improved version of SCAPE-M that has a correction in band 2 (Domínguez et al., 2011) (see below). SCAPE-M was developed and validated in several European water bodies (Guanter et al., 2010), and the SCAPE-M equation is as follows: LTOA ¼ L0 þ
1 rs Edir mil þ Edif T[ p 1 Srs
(1)
where LTOA represents the surface reflectance images derived from the top-of atmosphere radiance; L0 is the atmospheric path radiance; mil is the cosine of the illumination zenith angle qil measured between the solar ray and surface normal; Edirmil and Edif are the direct and diffuse fluxes, respectively, arriving at the surface; S is the atmospheric spherical albedo, which represents the reflectance of the atmosphere for isotropic light entering from the surface; T[ is the total atmospheric transmittance (for diffuse plus direct radiation) in the observation direction; and rs is the surface reflectance. SCAPE-M was corrected for the MERIS band 2 by means of an interpolation between the values of band 1 and band 3; if the original value is used, maximum values for vegetation and water are obtained because those values correspond to the absorption of Chl a. Thus, the correction with SCAPE-M is performed twice, and the assigned value of band 2
corresponds to the interpolated value of band 2 minus the value obtained without interpolation (Domínguez et al., 2011). The result of this improvement is the atmospheric correction algorithm SCAPE-M_B2. A water mask was used in the images to separate water zones from land areas, including the dam. Water/land pixels were discriminated with a simple threshold in the nearinfrared band (Agha et al., 2012). From each water pixels, the Chl a and PC values were estimated with the following algorithms (Domínguez et al., 2011) to generate MERIS thematic mappings: when Chl a 19:34 mg m3 : Chl a ¼ 19:34e6:1257
(2)
when Chl a 19:34 mg m3 : Chl a ¼ 19:34e5:2044½ðB9B7Þ=ðB9þB7Þ
(3)
PC ¼ 46: 478e5:186½ðB9B6Þ=ðB9þB6Þ
(4)
where B2 is the fluorescence of the first absorption peak of chlorophyll-a at approximately 442 nm (MERIS band 2); B5 is the maximum solar radiation that penetrates water at 560 nm and corresponds to the green band (MERIS band 5); B6 corresponds to the absorption peak of PC centred at 620 nm (MERIS band 6); B7 is the second absorption peak of chlorophyll-a at approximately 665 nm (MERIS band 7); and B9 is the fluorescence of chlorophyll-a at approximately 705 nm (MERIS band 9). The Chl a algorithms were calibrated with HPLC field data, and the PC algorithm was calibrated with a calibrating fluorometer (Domínguez et al., 2011). The Chl a and PC values retrieved from the MERIS thematic mapping correspond to the water pixels located at a midpoint near the dam, and they cover the points where field samples were obtained. The MERIS sensor captured the first optical thickness, which is equivalent to 0.6 times the Secchi disc depth.
2.4.
Data selection and area of study
Within the database, two conditions for image inclusion were established. 1) The MERIS image must have a clear pixel that covers the field sampling point. Consequently, water reservoirs with narrow dams or small surfaces were rejected. 2) Only field sampling data with cyanobacterial biovolume greater than 0.2 mm3/l are used. This threshold was selected because it corresponds to the first alert level for a cyanobacterial hazard. Using these conditions, we selected 23 different Spanish water reservoirs and 65 field samples (Fig. 1, Table 1). Twelve of the water reservoirs are in the Tajo river watershed, and 21 are in the west-central region of Spain, which is characterised by a hot-summer Mediterranean climate. Regions with this
Fig. 2 e Comparisons of chlorophyll-a concentrations between both methodologies. Both linear regressions were forced to begin at the origin (0,0) for validation purposes. The broken red line represents y ¼ x (RMSE ¼ root mean square error). a) Comparison of chlorophyll-a concentration measuring methodologies. n ¼ 52; p < 0.001; RMSE ¼ 0.024. b) Comparison of normalised chlorophyll-a concentration measuring methodologies. n ¼ 52; p < 0.001; RMSE ¼ 0.031. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
15
w a t e r r e s e a r c h 6 3 ( 2 0 1 4 ) 1 0 e2 0
Table 2 e Summary of the time differences between both methodologies (n ¼ 65).
Days of difference between the MERIS image and field sampling (absolute) Matching data information
Minimum
Maximum
Mean
Median
0
14
3.27
3
Matching sample dates
Satellite image retrieval not blocked by cloud cover (not including matching sample dates) 24
Total times that satellite image retrieval was blocked by cloud cover
Total times that satellite image retrieval was blocked by cloud cover more than once 6
8
43
it is the reservoir with the densest data as a result of having the most consecutive summer field samples available in the database (from 2007 to 2010).
climate typically experience hot, dry summers and mild, wet winters, and precipitation is higher during the colder months. Comparing the MERIS sensor images to the field data is especially useful in regions with climates characterised by infrequent cloud cover days because additional MERIS images would be available for the field sampling date.
3.
Results and discussion
2.5.
3.1.
Chlorophyll-a comparison
Results validation and statistical analysis
We validated the empirical results in the complete dataset using an external dataset from a hypertrophic water reservoir, the Rosarito reservoir. The Rosarito reservoir belongs to the Tietar River and is located in the Tajo watershed; the dam height is 38 m, and it has a water surface of 1475 km2 and a capacity of 92 hm3. The Rosarito reservoir was chosen because
Lab-measured and remote-sensing estimated Chl a concentrations (Fig. 2a) were compared, and a strong correlation between them was found (R2 ¼ 0.967; p < 0.001). Even when data are normalised, the correlation is still strong (R2 ¼ 0.947; p < 0.001) (Fig. 2b). Therefore, we can conclude that outliers do not produce strong leverage against the R2 value.
Fig. 3 e Correlation between phycocyanin concentration estimated by remote sensing and total cyanobacterial biovolume. n ¼ 65; p < 0.001; RMSE ¼ 0.06. (RMSE ¼ root mean square error).
16
w a t e r r e s e a r c h 6 3 ( 2 0 1 4 ) 1 0 e2 0
It is well known that the distribution of cyanobacteria in a water body is not perfectly homogeneous, and papers have reported that a patchy distribution of scum can produce unreliable pigment estimations by the MERIS sensor (Kutser, 2004). However, in the present work, no scum episodes were detected during the sampling periods. Therefore, no impact on the correlation was expected as a result of the heterogeneous distribution of scum. Another possible cause of error is the vertical distribution of the cyanobacterial community. Field samples were collected from the maximum Chl a layer within the euphotic zone, which has a maximum depth equal to 2.5 times the depth of the Secchi disk. However, the MERIS sensor covers the first optical thickness, which reaches 0.6 times the depth of the Secchi disk. Therefore, when most of the cyanobacterial community is located deeper than the first optical thickness, the community might not be detected by the MERIS sensor. However, when heterogeneity in the vertical cyanobacterial distribution is not intense, recent publications have found a good agreement between the MERIS sensor pigment estimation and field samples (Ruiz-Verdu´ et al., 2008; Guanter et al., 2010; Domínguez et al., 2011; Agha et al., 2012; Dekker and Hestir, 2012; Matthews et al., 2012; Mishra et al., 2013; Ali et al., 2014). In this work, field samples were collected at a certain distance from the dams where the distribution of the phytoplanktonic communities is generally expected to be more
homogenous (Kimmel et al., 1990). Moreover, the algorithms used to estimate both the Chl a and PC values in our work have been proven reliable and accurate compared to field samples (Domínguez et al., 2011). Major differences between the Chl a levels in the field and MERIS data were considered to be a result of heterogeneity in the cyanobacterial community, both at the temporal (Table 2) and spatial scales, and no relevant changes in the phytoplankton composition were expected given the short time lag between both measurements. Our data show a high correlation between the Chl a datasets (Fig. 2), and the MERIS data can be considered highly reflective of the field-sampled community.
3.2. Phycocyanin and cyanobacterial biovolume correlation The PC values estimated by remote sensing and the total cyanobacterial biovolume measured in the field samples (Fig. 3) were found to be significantly correlated (R2 ¼ 0.6219; p < 0.001), and the values are related by the following function: Cyanobacterial biovolume ¼ ð132:39$PCÞ 1825:1
(5)
Previous reports also found high correlations between the cyanobacterial biovolume and PC values estimated by waterborne radiometry analysis (Randolph et al., 2008).
Fig. 4 e Correlation between phycocyanin concentrations estimated by remote sensing and number of cells per millilitre. n ¼ 65; p < 0.001; RMSE ¼ 0.19. (RMSE ¼ root mean square error).
w a t e r r e s e a r c h 6 3 ( 2 0 1 4 ) 1 0 e2 0
The potential regression between the number of cells per volume and PC showed the strongest determination coefficient. However, it still shows a weak correlation(Fig. 4), although it was significant(R2 ¼ 0.362; p < 0.001). This relationship was slightly weaker than that found in other studies comparing the PC values estimated by remote sensing with cell counts (Hunter et al., 2010), and it is remarkably weaker than other reported field studies that used probe samples (Brient et al., 2008). However, this may have been a result of our study including 23 different water bodies with a heterogeneous species composition, which may scatter the data distribution. The phycobiliprotein cell quota can vary drastically for different cyanobacteria (Patel et al., 2005). Therefore, our results clearly identify biovolume as the most reliable biomass parameter to be estimated by remote sensing techniques in a diverse group of water bodies. When we considered the taxonomic distribution of the cyanobacteria found in the water bodies and subdivided the dataset by cyanobacterial order (Fig. 5), we found the best correlation between biovolume and PC for the order Oscillatoriales (R2 ¼ 0.726, p < 0.001). Nostocales and Chroococcales showed weaker correlations (Nostocales: R2 ¼ 0.422, p < 0.001; Chroococcales: R2 ¼ 0.517; p < 0.05). Although the correlation of Nostocales is the weakest of the three orders, the pattern described
17
by the linear regression and biovolume/pigment ratio is similar to that of Oscillatoriales and total cyanobacteria. The weaker correlation of Chroococcales and the different pattern followed by its biovolume/pigment correlation could be caused by the dominance of Microcystis aeruginosa (which was dominant in 6 of the 12 Chroococcales samples). Microcystis aeruginosa has been described as highly variable in terms of its cellular pigment content, even when grown in laboratory conditions (BanaresEspana et al., 2007). It is also important to note that Microcystis aeruginosa forms large 3D colonies that can exhibit unexpected optical behaviour (scattering within the colony) and can make cell counting difficult. The relatively low number of samples from water bodies dominated by Chroococcales (Chroococcales ¼ 12; Nostocales ¼ 30; Oscillatoriales ¼ 22) could also partly explain the weaker correlation. Considering the statistical robustness of our results, we can conclude that the equation given by the linear regression between the cyanobacterial biovolume and PC values estimated by remote sensing is a good method for calculating cyanobacterial biovolume using remote sensing tools. The dataset used in our study is diverse in terms of cyanobacterial abundance and taxonomic distribution, so our equation should be robust for assessing cyanobacterial biomass (as biovolume) in a wide range of ecosystems. This result represents an improvement
Fig. 5 e Correlation between phycocyanin concentrations and total cyanobacterial biovolume depending on the dominant order (RMSE ¼ root mean square error). (Nostocales: n ¼ 30, p < 0.001, RMSE ¼ 0.08; Oscillatoriales: n ¼ 30, p < 0.001, RMSE ¼ 0.07; Chroococcales: n ¼ 13, p < 0.05, RMSE ¼ 0.08).
18
w a t e r r e s e a r c h 6 3 ( 2 0 1 4 ) 1 0 e2 0
Fig. 6 e Time line comparing the estimated cyanobacterial biovolume (calculated from phycocyanin estimated by MERIS) and field sampled cyanobacterial biovolume.
over previously published methods, which were based on a limited number of water bodies and community types (Randolph et al., 2008; Hunter et al., 2010; Matthews et al., 2012).
3.3.
Validation
The equation obtained with the entire dataset was applied to the MERIS imagery of the Rosarito reservoir to obtain an
estimated cyanobacterial biovolume time series. This time series was compared to the biovolume data measured in an extensive sampling schedule spanning 4 years. The results (Fig. 6) show a high correlation between the biovolume measured in the field samples and the biovolume estimated by applying the equation to the satellite imagery. However, some discrepancies were found in 2010 that were most likely a result of a significant difference between the sampling dates and
Fig. 7 e Blue line (left y axis): determination coefficients (R2) of the correlation between the estimated cyanobacterial biovolume and field sampled cyanobacterial biovolume in the Rosarito reservoir depending on the days of difference between both date. Red broken line (right y axis): number of samples for each correlation. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
w a t e r r e s e a r c h 6 3 ( 2 0 1 4 ) 1 0 e2 0
image acquisition dates. In fact, when data are disaggregated by the time lag, the determination coefficients of the respective correlations between the field sampling and imaging show a clear decrease after a difference of 9 days between the acquisition of the data with both methodologies (Fig. 7). It is obvious that when sampling and imaging occur on the same day, the correlation is very good (R2 ¼ 0.72; p < 0.001). When all of the time lags are considered, the determination coefficients are weaker, but they are still significant until a difference of 9 days (R2 varies between 0.51 and 0.55; p < 0.001). However, when the time lag spans longer than 9 days, there is no significant correlation (R2 ¼ 0.1676; p ¼ 0.06). Although the MERIS sensor is now inoperative, Sentinel-3 is expected to have a similar sensor available that will also retrieve L1b products after 1 h of scanning (Donlon et al., 2012). Therefore, the Sentinel-3 mission might be a useful monitoring tool in support of management strategies for water bodies larger than 9 ha, especially for countries with few cloudy days. The previously recorded MERIS images can also represent a useful data source for understanding cyanobacterial dynamics in certain water bodies over long-term periods; in certain cases, up to 10 years of high frequency data can become crucial to the development of management strategies and increased forecasting capabilities.
4.
Conclusions
In this work, we define and validate the use of MERIS imagery to estimate the cyanobacterial biovolume in freshwater ecosystems under non-scum conditions. This approach is based on the detection of the cyanobacterial pigment PC by means of the MERIS sensor, and we have adapted this tool to calculate cyanobacterial biovolume, an important parameter for limnological and water quality studies. In our analysis, the correlation between the measured cyanobacterial biovolume and PC values estimated from the MERIS imagery was found to be a robust proxy for cyanobacterial biomass estimation. The regression equation was built by using data from 23 aquatic ecosystems in the Iberian Peninsula that were dominated by diverse cyanobacterial communities. A validation exercise conducted with the time series data from an external water reservoir indicated that the equation was useful for estimating cyanobacterial biomass and that it had excellent results when the time lag between the field measurements and MERIS imaging was less than 9 days. Considering that our data covered a wide area of different lithological and trophic characteristics, included 89 different cyanobacteria species (of which 23 different species were dominant) and produced positive results from the validation test, we believe that these results could be transferable to other water reservoirs.
Acknowledgements This research was funded by a grant from the Centre for Studies and Experimentation of Public Works (CEDEX) of the Urbanism Ministry of Spain. Special thanks to the Centre for
19
Hydrographic Studies and everyone who helped to obtain the data from the water reservoirs for this research.
Appendix A. Supplementary data Supplementary data related to this article can be found at http://dx.doi.org/10.1016/j.watres.2014.06.001.
references
s, S., Wo € rmer, L., Domínguez, J.A., Quesada, A., 2012. Agha, R., Cire Multi-scale strategies for the monitoring of freshwater cyanobacteria: reducing the sources of uncertainty. Water Res. 46 (9), 3043e3053. Ali, K., Witter, D., Ortiz, J., 2014. Application of empirical and semi-analytical algorithms to MERIS data for estimating chlorophyll a in case 2 waters of Lake Erie. Environ. Earth Sci. 71 (9), 4209e4220. Banares-Espana, E., Lopez-Rodas, V., Costas, E., Salgado, C., Flores-Moya, A., 2007. Genetic variability associated with photosynthetic pigment concentration, and photochemical and nonphotochemical quenching, in strains of the cyanobacterium Microcystis aeruginosa. FEMS Microbiol. Ecol. 60 (3), 449e455. Binding, C.E., Greenberg, T.A., Bukata, R.P., 2011. Time series analysis of algal blooms in lake of the woods using the MERIS maximum chlorophyll index. J. Plankton Res. 33 (12), 1847e1852. Brient, L., Lengronne, M., Bertrand, E., Rolland, D., Sipel, A., Steinmann, D., Baudin, I., Legeas, M., Le Rouzic, B., Bormans, M., 2008. A phycocyanin probe as a tool for monitoring cyanobacteria in freshwater bodies. J. Environ. Monit. 10 (2), 248e255. Carvalho, L., McDonald, C., de Hoyos, C., Mischke, U., Phillips, G., Borics, G., Poikane, S., Skjelbred, B., Solheim, A.L., Van Wichelen, J., Cardoso, A.C., 2013a. Sustaining recreational quality of European lakes: minimizing the health risks from algal blooms through phosphorus control. J. Appl. Ecol. 50 (2), 315e323. Carvalho, L., Poikane, S., Solheim, A.L., Phillips, G., Borics, G., Catalan, J., De Hoyos, C., Drakare, S., Dudley, B.J., Jarvinen, M., Laplace-Treyture, C., Maileht, K., McDonald, C., Mischke, U., Moe, J., Morabito, G., Noges, P., Noges, T., Ott, I., Pasztaleniec, A., Skjelbred, B., Thackeray, S.J., 2013b. Strength and uncertainty of phytoplankton metrics for assessing eutrophication impacts in lakes. Hydrobiologia 704 (1), 127e140. Chorus, I., 2012. Current Approaches to Cyanotoxin Risk Assessment, Risk Management and Regulations in Different Countries. Federal Environment Agency, Berlin. Chorus, I., Bartram, J., 1999. Toxic Cyanobacteria in Water. Spon Press. De Hoyos, C., N., A.I., Aldasoro, J.J., 2004. Cyanobacteria distribution and abundance in the Spanish water reservoirs during thermal stratification. Limnetica 23 (1e2), 119e132. Dekker, A.G., Hestir, E.L., 2012. Evaluating the Feasibility of Systematic Inland Water Quality Monitoring with Satellite Remote Sensing. Domínguez, J.A., Alonso, C., Alonso, A., 2011. Remote sensing as a tool for monitoring water quality parameters for Mediterranean Lakes of European Union water framework directive (WFD) and as a system of surveillance of cyanobacterial harmful algae blooms (SCyanoHABs). Environ. Monit. Assess. 181 (1e4), 317e334.
20
w a t e r r e s e a r c h 6 3 ( 2 0 1 4 ) 1 0 e2 0
Donlon, C., Berruti, B., Buongiorno, A., Ferreira, M.H., Femenias, P., Frerick, J., Goryl, P., Klein, U., Laur, H., Mavrocordatos, C., Nieke, J., Rebhan, H., Seitz, B., Stroede, J., Sciarra, R., 2012. The global monitoring for environment and security (GMES) sentinel-3 mission. Remote Sens. Environ. 120, 37e57. Guanter, L., Ruiz-Verdu, A., Odermatt, D., Giardino, C., Simis, S., Estelles, V., Heege, T., Dominguez-Gomez, J.A., Moreno, J., 2010. Atmospheric correction of ENVISAT/MERIS data over inland waters: validation for European lakes. Remote Sens. Environ. 114 (3), 467e480. Hart, B.T., 1984. Eutrophication of waters e monitoring, assessment and control e OECD. Search 15 (1e2), 50. Hillebrand, H., Du¨rselen, C.-D., Kirschtel, D., Pollingher, U., Zohary, T., 1999. Biovolume calculation for pelagic and benthic microalgae. J. Phycol. 35 (2), 403e424. Hunter, P.D., Tyler, A.N., Carvalho, L., Codd, G.A., Maberly, S.C., 2010. Hyperspectral remote sensing of cyanobacterial pigments as indicators for cell populations and toxins in eutrophic lakes. Remote Sens. Environ. 114, 2705e2718. Jiang, T., Zhang, J.P., Chang, W.R., Liang, D.C., 2001. Crystal structure of R-phycocyanin and possible energy transfer pathways in the phycobilisome. Biophys. J. 81 (2), 1171e1179. Kimmel, B.L., Lind, O.T., Paulson, L.J., 1990. Reservoir primary production. In: Thorton, K.W., Kimmel, B.L., Payne, F.E. (Eds.), Reservoir Limnology: Ecological Perspectives. John Wiley & Sons, Inc, pp. 133e193. Kutser, T., 2004. Quantitative detection of chlorophyll in cyanobacterial blooms by satellite remote sensing. Limnol. Oceanogr. 49 (6), 2179e2189. Matthews, M.W., Bernard, S., Robertson, L., 2012. An algorithm for detecting trophic status (chlorophyll-a), cyanobacterialdominance, surface scums and floating vegetation in inland and coastal waters. Remote Sens. Environ. 124, 637e652. Mishra, S., Mishra, D.R., Lee, Z., Tucker, C.S., 2013. Quantifying cyanobacterial phycocyanin concentration in turbid productive waters: a quasi-analytical approach. Remote Sens. Environ. 133, 141e151. Parsons, T.R., Strickland, J.D.H., 1968. A Practical Handbook of Seawater Analysis. Pigment analysis. Otawa. Supply and Services Canada. Patel, A., Mishra, S., Pawar, R., Ghosh, P.K., 2005. Purification and characterization of C-phycocyanin from cyanobacterial species of marine and freshwater habitat. Protein Expr. Purif. 40 (2), 248e255.
Quesada, A., Sanchis, D., Carrasco, D., 2004. Cyanobacteria in Spanish reservoirs. How frequently are toxic? Limnetica 23 (1e2), 109e118. Randolph, K., Wilson, J., Tedesco, L., Li, L., Pascual, D.L., Soyeux, E., 2008. Hyperspectral remote sensing of cyanobacteria in turbid productive water using optically active pigments, chlorophyll a and phycocyanin. Remote Sens. Environ. 112, 4009e4019. Rast, M., Bezy, J.L., Bruzzi, S., 1999. The ESA medium resolution imaging spectrometer MERIS e a review of the instrument and its mission. Int. J. Remote Sens. 20 (9), 1681e1702. Rowan, K.S., 1989. Photosynthetics Pigments of Algae. Cambridge University Press, Cambridge. ~ aRuiz-Verdu´, A., Simis, S.G.H., de Hoyos, C., Gons, H.J., Pen Martínez, R., 2008. An evaluation of algorithms for the remote sensing of cyanobacterial biomass. Remote Sens. Environ. 112 (11), 3996e4008. Simis, S.G.H., Ruiz-Verdu, A., Dominguez-Gomez, J.A., PenaMartinez, R., Peters, S.W.M., Gons, H.J., 2007. Influence of phytoplankton pigment composition on remote sensing of cyanobacterial biomass. Remote Sens. Environ. 106 (4), 414e427. Sournia, A., 1978. Phytoplankton Manual, vol. 1. Paris. Spoerke, D.G., Rumack, B.H., 1985. Blue-green algae poisoning. J. Emerg. Med. 2 (5), 353e355. Svircev, Z., Simeunovic, J., Subakov-Simic, G., Krstic, S., Pantelic, D., Dulic, T., 2013. Cyanobacterial blooms and their toxicity in Vojvodina lakes, Serbia. Int. J. Environ. Res. 7 (3), 845e858. Vasconcelos, V., 2006. Eutrophication, toxic cyanobacteria and cyanotoxins: when ecosystems cry for help. Limnetica 25 (1e2), 425e432. Whitton, B.A., Potts, M., 2002. Their Diversity in Time and Space, vol. 1. Springer, Netherlands. Willen, E., 1997. Phytoplankton in water quality assessment, an indicator concept. Phycologia 36 (4), 123e124.
url's
http://eur-lex.europa.eu/LexUriServ/LexUriServ.do?uri¼CELEX: 32000L0060:en:HTML. https://earth.esa.int/web/guest/missions/esa-operational-eomissions/envisat/instruments/meris/design. https://earth.esa.int/web/guest/missions/esa-future-missions/ sentinel-3.